Our world is constantly measured where phenomena are often indexed in space which is then referred to as spatial data. The amount of data collected this way is huge and it is thought that not all that data is informative and the working premise of dimension reduction is that all relevant information lies in a signal subspace and the rest of the space contains only noise. The goal in this project is to derive dimension reduction methods for such data which are based only on the information from the random phenomena, which means they are blind and therefore known as blind source separation (BSS) methods. The phenomena under consideration can for example be vectors like geochemical compositions of soil at different locations or functions as for example chemical spectra obtained from soil samples at these locations. Usually the dimension of the signal space is unknown and tools for its estimation are to be developed too, together with efficient BSS software.

2024

2028